Unverified50% confidenceBenchmarkExact time
Under extreme weights, the dispersion of the estimates from pooled targeting is 1.7–1.8 times that of fold-wise targeting.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe current conclusion comes from only a single source, and whether the ratio of 1.7 points to 2.1 times cost holds stable across different task distributions still requires more independent replication.68% similarUnverifiedIn the specific experimental case, the overlapping-anchor scheme (n=912) showed the model at 55.2% versus the majority class at 51.1%, a net advantage of +4.1pp, about 2.4 standard errors; the non-overlapping-anchor scheme (n=163) showed 59.5% versus 54.0%, a net advantage of +5.5pp, about 1.4 standard errors.68% similarUnverifiedSubstituting the strategy's data (55% × 2.4R − 45% × 1R) yields an expected value of +0.87R, meaning a long-term average net gain of 0.87 units for every 1 unit of risk taken.65% similarUnverifiedUnder the DeepSuite V1.1 benchmark, the same model's scores fluctuate from 69.8% to 74.2% across different harnesses.65% similarUnverified靶向数据工程中，针对性数据占总微调数据的 5%-15% 通常能取得良好效果，过高比例可能导致模型在其他维度退化65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023534API
curl https://kongchang.com/api/v1/knowledge/claims/1023534MCP
get_claim(id=1023534)